Fix RARE2012, add RARE2007

CHANGELOG.md

@@ -18,6 +18,10 @@
   * In general, naming convention for attriutes has been changed to use the plural form if the attribute is a list of values for each
     element (i.e., `Scanpaths.xs`) and the singular form if the attribute is a single value (i.e., `Scanpaths.length`,  `Fixations.x`). This resulted in
     renaming `Fixations.subjects` to `Fixations.subject`. The old name is now a deprecated alias.
+  * Bugfix: Compatibility with torch 2.2 and numpy 2.0
+  * Bugfix!: The download location of the RARE2012 model changed. The new source code results in slightly different predictions.
+  * Feature: The RARE2007 model is now available as `pysaliency.external_models.RARE2007`. It's execution requires MATLAB.
+
 
 * 0.2.22:
   * Enhancement: New [Tutorial](notebooks/Tutorial.ipynb).

pysaliency/external_models/__init__.py

@@ -1,19 +1,18 @@
+from .deepgaze import (
+    DeepGazeI,
+    DeepGazeIIE,
+)
 from .matlab_models import (
     AIM,
-    SUN,
-    ContextAwareSaliency,
     BMS,
     GBVS,
-    GBVSIttiKoch,
-    Judd,
-    IttiKoch,
+    RARE2007,
     RARE2012,
+    SUN,
+    ContextAwareSaliency,
     CovSal,
+    GBVSIttiKoch,
+    IttiKoch,
+    Judd,
 )
-
-from .deepgaze import (
-    DeepGazeI,
-    DeepGazeIIE,
-)
-
 from .utils import ExternalModelMixin

pysaliency/external_models/matlab_models.py

@@ -454,18 +454,61 @@ def _setup(self, saliency_toolbox_archive):
             f.write(resource_string(__name__, 'scripts/IttiKoch_wrapper.m'))
 
 
+class RARE2007(ExternalModelMixin, MatlabSaliencyMapModel):
+    """
+    RARE2007 by Mancas.
+    The original matlab code is used.
+
+    .. note::
+        The source code published by the authors is used without modifications.
+
+    .. seealso::
+        Matei Mancas. Relative influence of bottom-up and top-down attention.
+        [International Workshop on Attention in Cognitive Systems 2008]
+
+        https://numediart.github.io/VisualAttention-RareFamily/
+        https://github.com/numediart/VisualAttention-Rare2007
+    """
+    __modelname__ = 'RARE2007'
+
+    def __init__(self, location=None, **kwargs):
+        self.setup(location)
+        super(RARE2007, self).__init__(os.path.join(self.location, 'RARE2007_wrapper.m'), only_color_stimuli=True, **kwargs)
+
+    def _setup(self):
+        source_location = os.path.join(self.location, 'source')
+        directory = 'VisualAttention-Rare2007-ee7ef25c9b6cc7e96837091e868d03bd97b4b131'
+        print('Downloading RARE2007 Model...')
+        url = 'https://github.com/numediart/VisualAttention-Rare2007/archive/ee7ef25c9b6cc7e96837091e868d03bd97b4b131.zip'
+        download_extract_patch(url,
+                               '7ee05ae059dac10478b5cabc8ee066dc',
+                               os.path.join(source_location, directory),
+                               location_in_archive=True,
+                               patches=None)
+
+        with open(os.path.join(self.location, 'RARE2007_wrapper.m'), 'wb') as f:
+            f.write(resource_string(__name__, 'scripts/RARE2007_wrapper.m'))
+
+
 class RARE2012(ExternalModelMixin, MatlabSaliencyMapModel):
     """
     RARE2012 by Margolin et al.
     The original matlab code is used.
 
     .. note::
-        The original code is patched to work from other directories.
 
-        The model makes use of the [SaliencyToolbox](http://www.saliencytoolbox.net/). Due
-        to licence restrictions the Toolbox cannot be downloaded automatically. You have to
-        download it yourself and provide the location of the zipfile via the
-        `saliency_toolbox_archive`-keyword to the constructor.
+
+        This page is not available anymore, the RARE models are now hosted
+        at https://numediart.github.io/VisualAttention-RareFamily/, the RARE2012
+        model is published at https://github.com/numediart/VisualAttention-Rare2012.
+        This model is a wrapper around the published (compiled) matlab code.
+
+        The model used to be hosted at
+        http://www.tcts.fpms.ac.be/attention/?categorie16/what-and-why.
+        This page is not available anymore, the RARE models are now hosted
+        at https://numediart.github.io/VisualAttention-RareFamily/ and GitHub.
+        With this change, the behaviour of RARE2012 changed slightly
+        compared to the earlier version (e.g., saliency maps look a bit smoother).
 
         This model does not work with octave due to incompabilities
         of octave with matlab (RARE2012 comes as precompiled matlab file).
@@ -477,7 +520,7 @@ class RARE2012(ExternalModelMixin, MatlabSaliencyMapModel):
         saliency detection with its comparative statistical analysis
         [Signal Processing: Image Communication, 2013]
 
-        http://www.tcts.fpms.ac.be/attention/?categorie16/what-and-why
+        https://numediart.github.io/VisualAttention-RareFamily/
     """
     __modelname__ = 'RARE2012'
 
@@ -487,18 +530,12 @@ def __init__(self, location=None, **kwargs):
 
     def _setup(self):
         source_location = os.path.join(self.location, 'source')
+        directory = 'VisualAttention-Rare2012-55ba7414b971429e5e899ddfa574e4235fc806e6'
         print('Downloading RARE2012 Model...')
-        #download_extract_patch('http://www.tcts.fpms.ac.be/attention/data/documents/data/rare2012.zip',
-        download_extract_patch('http://tcts.fpms.ac.be/attention/data/medias/documents/models/rare2012.zip',
-                               '5a0a0de83e82b46fa70cea8b0a6bae55',
-                               os.path.join(source_location, 'Rare2012'),
-                               location_in_archive=True,
-                               patches=None)
-
-        print('Downloading simplegabortb-v1.0.0')
-        download_extract_patch('http://www2.it.lut.fi/project/simplegabor/downloads/src/simplegabortb/simplegabortb-v1.0.0.tar.gz',
-                               '92bc6ae7178a7b1301fad52489f5d677',
-                               os.path.join(source_location, 'simplegabortb-v1.0.0'),
+        url = 'https://github.com/numediart/VisualAttention-Rare2012/archive/55ba7414b971429e5e899ddfa574e4235fc806e6.zip'
+        download_extract_patch(url,
+                               'a5bbd8a42c231530834659acfd0d2b64.',
+                               os.path.join(source_location, directory),
                                location_in_archive=True,
                                patches=None)
 

pysaliency/external_models/scripts/RARE2012_wrapper.m

@@ -1,9 +1,8 @@
 function [ ] =  RARE2012(filename, outname)
 
-    addpath('source/simplegabortb-v1.0.0')
-    addpath('source/Rare2012')
+    addpath('source/VisualAttention-Rare2012-55ba7414b971429e5e899ddfa574e4235fc806e6')
 
 	I = im2double(imread(filename));
-    saliency_map = RARE2012(I);
+    saliency_map = rare2012(I);
     save(outname, 'saliency_map');
-    
+

tests/conftest.py

@@ -49,20 +49,14 @@ def matlab(request, pytestconfig):
         pysaliency.utils.MatlabOptions.octave_names = []
     elif request.param == 'octave':
         if pytestconfig.getoption("--nooctave"):
-            pytest.skip("skipped octave")
+            pytest.skip("skipped octave due to command line option")
+        elif any([marker.name == 'skip_octave' for marker in request.node.own_markers]):
+            pytest.skip("skipped octave due to test marker")
         pysaliency.utils.MatlabOptions.matlab_names = []
         pysaliency.utils.MatlabOptions.octave_names = ['octave', 'octave.exe']
 
     return request.param
 
-
-@pytest.fixture()
-def skip_by_matlab(request, matlab):
-    if request.node.get_marker('skip_octave'):
-        if matlab == 'octave':
-            pytest.skip('skipped octave')
-
-
 #@pytest.fixture(params=["no_location", "with_location"])
 #def location(tmpdir, request):
 #    if request.param == 'no_location':

tests/test_external_models.py

@@ -120,6 +120,20 @@ def test_IttiKoch(tmpdir, matlab, color_stimulus, grayscale_stimulus):
                                np.load(os.path.join('tests', 'external_models', '{}_grayscale_stimulus.npy'.format(model.__modelname__))))
 
 
+@pytest.mark.skip_octave
+@pytest.mark.matlab
+def test_RARE2007(tmpdir, matlab, color_stimulus, grayscale_stimulus):
+    model = pysaliency.external_models.RARE2007(location=str(tmpdir))
+    print('Testing color')
+    saliency_map = model.saliency_map(color_stimulus)
+    np.testing.assert_allclose(saliency_map,
+                               np.load(os.path.join('tests', 'external_models', '{}_color_stimulus.npy'.format(model.__modelname__))))
+    print('Testing Grayscale')
+    saliency_map = model.saliency_map(grayscale_stimulus)
+    np.testing.assert_allclose(saliency_map,
+                               np.load(os.path.join('tests', 'external_models', '{}_grayscale_stimulus.npy'.format(model.__modelname__))))
+
+
 @pytest.mark.skip_octave
 @pytest.mark.matlab
 def test_RARE2012(tmpdir, matlab, color_stimulus, grayscale_stimulus):